Modeling Accurate Prediction of Junctionless Infrared Phototransistor Performance Using Explainable DL Regression Models

Abstract

In this study, a deep learning (DL)-guided optimization strategy is developed to improve the optoelectronic performance of planar junctionless phototransistors. For this purpose, we have opted to combine accurate two-dimensional (2D) analytical and numerical simulations with advanced DL approaches to investigate key design parameters impact on device performance. So, we have simulated the current-voltage (I-V) characteristics of JLP devices under infrared illumination conditions and thoroughly analyze the influences of material property variations, channel doping levels, and device geometry on the phototransistor key performance metrics. And, diverse DL algorithms, including deep learning classification models, are used to evaluate and parameters significance. As well, feature importance analysis is performed to identify the key factors influencing the overall device responsivity (R). The results show that optimized doping and bias conditions yield a high responsivity with a reduced dark current. They reveal key design parameters that significantly impact phototransistor efficiency and provide valuable insights for optimizing device design. Furthermore, these findings highlight that the synergy between device physics and DL-based modeling provides a powerful methodology for the design of next-generation low-power, low-cost, and high-performance junctionless phototransistors (JLPs) for inter-chip data communication applications.